Preserve nonbreaking spaces and unit symbols visibly

Preserve characters before making them convenient

Preserve the original text and inspect deliberate conversions so cleanup does not change the evidence you need to retain.

Before paying for technical-document extraction, inspect how the source meaning survives in the output. ParseMill’s public material-register example gives you a synthetic PDF and downloadable evidence to compare. Bring your own document type and acceptance checks to a scope discussion; the self-service invoice plan is a separate offer.

Decide whether the goal is faithful transcription, consistent display or machine matching. Each may need a different field. Do not call a normalized display string the original source text.

Inspect a synthetic character set

The examples are synthetic and their visual similarity is intentional. Character inspection can reveal what the extracted text contains, while comparison with the PDF establishes whether that text is faithful. A micro symbol and Greek mu may be normalized for a specific matching task only under an explicit rule.

Unicode: Normalization forms distinguishes canonical and compatibility normalization, including changes involving spacing and superscripts. Choose a form deliberately because compatibility changes can remove distinctions relevant to technical text.

Create a character inspection record

1. Copy the raw string into a text-preserving field and retain the source location. Compare it with the rendered PDF at useful magnification.
2. Inspect suspicious characters by code point using a trusted local text tool. Record the sequence when visual comparison alone is insufficient.
3. Create a separate normalized display or matching field with a documented transformation rule. Keep unit interpretation separate from character replacement.
4. Export and reopen the file in the receiving application. Check visible glyphs, exact raw strings and any matching keys independently.

NIST: SI notation checklist describes unit notation including spacing and exponent structure. Preserve those relationships instead of stripping all non-ASCII characters from a unit.

Handle missing glyphs and lookalikes

A square replacement glyph in a spreadsheet can be a font-display problem or actual character corruption. Compare the stored string with the original before deciding which. If the raw bytes or characters changed, recover from the source; changing the font alone will not restore them.

Avoid a blanket rule that replaces every superscript with an ordinary digit and removes punctuation. That can make square millimeters look like a different unit or collapse multiplication and division structure.

Check fidelity and convenience separately

Acceptance means the raw field remains recoverable and the normalized field follows its declared rule. The recipient should know which field is suitable for source review and which is intended for searching or matching. Keep unresolved unit interpretations visible.

Bring the unit typography in your own reports to a scoped document-workflow discussion. Agree which characters and source strings the output must preserve, then inspect them in the receiving spreadsheet during any pilot. General scientific-unit handling is not established by the invoice trial.